The Open-Source AI Stack
RSS
All models

Models · qwen3-5

Qwen3.5 Plus

Open Alibaba · 2026-02-16 · Apache-2.0

Flagship of the Qwen 3.5 generation. 397B MoE activating 17B per token with 512 experts. First Qwen open-weights model with native vision input. Per Alibaba's evals, the 397B-A17B outperforms the trillion-parameter Qwen3 Max on several reasoning and coding benchmarks while running roughly 19x faster at 256K context.

Architecture

tokens in Embedding vocab not disclosed · qwen tokenizer × N layers Attention (not disclosed) RoPE + YaRN context 1,000,000 tokens MoE Router 512 experts total · ? active per token shown: 32 of 512 Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
moe
Total params
397B
Active params
17B
Experts
512 total · ? active
Context window
1.0M tokens
Attention
unknown
Position encoding
rope-yarn
Post-training
sft, rlhf
OSI-approved
yes
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

Available quantizations

— not catalogued

Notable innovations

  • · 397B / 17B-active MoE with 512 experts
  • · First open-weights Qwen with native vision
  • · 1M-context hosted variant

Lineage

Qwen3.5-Plus is the hosted API variant of Qwen3.5-397B-A17B (not catalogued); the 397B MoE flagship of the 3.5 line, replacing the dense Qwen3 235B as the open headliner.

Sources